src/lib/audio_analyser.cc

   1 /*
   2     Copyright (C) 2021 Carl Hetherington <cth@carlh.net>
   3
   4     This file is part of DCP-o-matic.
   5
   6     DCP-o-matic is free software; you can redistribute it and/or modify
   7     it under the terms of the GNU General Public License as published by
   8     the Free Software Foundation; either version 2 of the License, or
   9     (at your option) any later version.
  10
  11     DCP-o-matic is distributed in the hope that it will be useful,
  12     but WITHOUT ANY WARRANTY; without even the implied warranty of
  13     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14     GNU General Public License for more details.
  15
  16     You should have received a copy of the GNU General Public License
  17     along with DCP-o-matic.  If not, see <http://www.gnu.org/licenses/>.
  18
  19 */
  20
  21
  22 #include "audio_analyser.h"
  23 #include "audio_analysis.h"
  24 #include "audio_buffers.h"
  25 #include "audio_content.h"
  26 #include "audio_filter_graph.h"
  27 #include "audio_point.h"
  28 #include "config.h"
  29 #include "dcpomatic_log.h"
  30 #include "film.h"
  31 #include "filter.h"
  32 #include "playlist.h"
  33 #include <dcp/warnings.h>
  34 extern "C" {
  35 #include <leqm_nrt.h>
  36 LIBDCP_DISABLE_WARNINGS
  37 #include <libavutil/channel_layout.h>
  38 #ifdef DCPOMATIC_HAVE_EBUR128_PATCHED_FFMPEG
  39 #include <libavfilter/f_ebur128.h>
  40 #endif
  41 LIBDCP_ENABLE_WARNINGS
  42 }
  43
  44
  45 using std::make_shared;
  46 using std::max;
  47 using std::shared_ptr;
  48 using std::vector;
  49 using namespace dcpomatic;
  50
  51
  52 static auto constexpr num_points = 1024;
  53
  54
  55 AudioAnalyser::AudioAnalyser(shared_ptr<const Film> film, shared_ptr<const Playlist> playlist, bool whole_film, std::function<void (float)> set_progress)
  56         : _film (film)
  57         , _playlist (playlist)
  58         , _set_progress (set_progress)
  59 #ifdef DCPOMATIC_HAVE_EBUR128_PATCHED_FFMPEG
  60         , _ebur128(film->audio_frame_rate(), film->audio_channels())
  61 #endif
  62         , _sample_peak (film->audio_channels())
  63         , _sample_peak_frame (film->audio_channels())
  64         , _analysis (film->audio_channels())
  65 {
  66
  67 #ifdef DCPOMATIC_HAVE_EBUR128_PATCHED_FFMPEG
  68         _filters.push_back({"ebur128", "ebur128", "audio", "ebur128=peak=true"});
  69         _ebur128.setup(_filters);
  70 #endif
  71
  72         _current = std::vector<AudioPoint>(_film->audio_channels());
  73
  74         if (!whole_film) {
  75                 _start = _playlist->start().get_value_or(DCPTime());
  76         }
  77
  78         for (int i = 0; i < film->audio_channels(); ++i) {
  79                 _sample_peak[i] = 0;
  80                 _sample_peak_frame[i] = 0;
  81         }
  82
  83         auto add_if_required = [](vector<double>& v, size_t i, double db) {
  84                 if (v.size() > i) {
  85                         v[i] = pow(10, db / 20);
  86                 }
  87         };
  88
  89         auto content = _playlist->content();
  90         if (whole_film) {
  91                 _leqm_channels = film->audio_channels();
  92         } else {
  93                 _leqm_channels = 0;
  94                 for (auto channel: content[0]->audio->mapping().mapped_output_channels()) {
  95                         /* This means that if, for example, a file only maps C we will
  96                          * calculate LEQ(m) for L, R and C.  I'm not sure if this is
  97                          * right or not.
  98                          */
  99                         _leqm_channels = std::min(film->audio_channels(), channel + 1);
 100                 }
 101         }
 102
 103         /* XXX: is this right?  Especially for more than 5.1? */
 104         vector<double> channel_corrections(_leqm_channels, 1);
 105         add_if_required (channel_corrections,  4,   -3); // Ls
 106         add_if_required (channel_corrections,  5,   -3); // Rs
 107         add_if_required (channel_corrections,  6, -144); // HI
 108         add_if_required (channel_corrections,  7, -144); // VI
 109         add_if_required (channel_corrections,  8,   -3); // Lc
 110         add_if_required (channel_corrections,  9,   -3); // Rc
 111         add_if_required (channel_corrections, 10,   -3); // Lc
 112         add_if_required (channel_corrections, 11,   -3); // Rc
 113         add_if_required (channel_corrections, 12, -144); // DBox
 114         add_if_required (channel_corrections, 13, -144); // Sync
 115         add_if_required (channel_corrections, 14, -144); // Sign Language
 116         add_if_required (channel_corrections, 15, -144); // Unused
 117
 118         _leqm.reset(new leqm_nrt::Calculator(
 119                 _leqm_channels,
 120                 film->audio_frame_rate(),
 121                 24,
 122                 channel_corrections,
 123                 850, // suggested by leqm_nrt CLI source
 124                 64,  // suggested by leqm_nrt CLI source
 125                 boost::thread::hardware_concurrency()
 126                 ));
 127
 128         DCPTime const length = _playlist->length (_film);
 129
 130         Frame const len = DCPTime (length - _start).frames_round (film->audio_frame_rate());
 131         _samples_per_point = max (int64_t (1), len / num_points);
 132 }
 133
 134
 135 void
 136 AudioAnalyser::analyse (shared_ptr<AudioBuffers> b, DCPTime time)
 137 {
 138         LOG_DEBUG_AUDIO_ANALYSIS("AudioAnalyser received %1 frames at %2", b->frames(), to_string(time));
 139         DCPOMATIC_ASSERT (time >= _start);
 140         /* In bug #2364 we had a lot of frames arriving here (~47s worth) which
 141          * caused an OOM error on Windows.  Check for the number of frames being
 142          * reasonable here to make sure we catch this if it happens again.
 143          */
 144         DCPOMATIC_ASSERT(b->frames() < 480000);
 145
 146 #ifdef DCPOMATIC_HAVE_EBUR128_PATCHED_FFMPEG
 147         if (Config::instance()->analyse_ebur128 ()) {
 148                 _ebur128.process(b);
 149         }
 150 #endif
 151
 152         int const frames = b->frames ();
 153         vector<double> interleaved(frames * _leqm_channels);
 154
 155         for (int j = 0; j < _leqm_channels; ++j) {
 156                 float const* data = b->data(j);
 157                 for (int i = 0; i < frames; ++i) {
 158                         float s = data[i];
 159
 160                         interleaved[i * _leqm_channels + j] = s;
 161
 162                         float as = fabsf (s);
 163                         if (as < 10e-7) {
 164                                 /* We may struggle to serialise and recover inf or -inf, so prevent such
 165                                    values by replacing with this (140dB down) */
 166                                 s = as = 10e-7;
 167                         }
 168                         _current[j][AudioPoint::RMS] += pow (s, 2);
 169                         _current[j][AudioPoint::PEAK] = max (_current[j][AudioPoint::PEAK], as);
 170
 171                         if (as > _sample_peak[j]) {
 172                                 _sample_peak[j] = as;
 173                                 _sample_peak_frame[j] = _done + i;
 174                         }
 175
 176                         if (((_done + i) % _samples_per_point) == 0) {
 177                                 _current[j][AudioPoint::RMS] = sqrt (_current[j][AudioPoint::RMS] / _samples_per_point);
 178                                 _analysis.add_point (j, _current[j]);
 179                                 _current[j] = AudioPoint ();
 180                         }
 181                 }
 182         }
 183
 184         _leqm->add(interleaved);
 185
 186         _done += frames;
 187
 188         DCPTime const length = _playlist->length (_film);
 189         _set_progress ((time.seconds() - _start.seconds()) / (length.seconds() - _start.seconds()));
 190         LOG_DEBUG_AUDIO_ANALYSIS_NC("Frames processed");
 191 }
 192
 193
 194 void
 195 AudioAnalyser::finish ()
 196 {
 197         vector<AudioAnalysis::PeakTime> sample_peak;
 198         for (int i = 0; i < _film->audio_channels(); ++i) {
 199                 sample_peak.push_back (
 200                         AudioAnalysis::PeakTime (_sample_peak[i], DCPTime::from_frames (_sample_peak_frame[i], _film->audio_frame_rate ()))
 201                         );
 202         }
 203         _analysis.set_sample_peak (sample_peak);
 204
 205 #ifdef DCPOMATIC_HAVE_EBUR128_PATCHED_FFMPEG
 206         if (Config::instance()->analyse_ebur128 ()) {
 207                 void* eb = _ebur128.get("Parsed_ebur128_0")->priv;
 208                 vector<float> true_peak;
 209                 for (int i = 0; i < _film->audio_channels(); ++i) {
 210                         true_peak.push_back (av_ebur128_get_true_peaks(eb)[i]);
 211                 }
 212                 _analysis.set_true_peak (true_peak);
 213                 _analysis.set_integrated_loudness (av_ebur128_get_integrated_loudness(eb));
 214                 _analysis.set_loudness_range (av_ebur128_get_loudness_range(eb));
 215         }
 216 #endif
 217
 218         if (_playlist->content().size() == 1) {
 219                 /* If there was only one piece of content in this analysis we may later need to know what its
 220                    gain was when we analysed it.
 221                 */
 222                 if (auto ac = _playlist->content().front()->audio) {
 223                         _analysis.set_analysis_gain (ac->gain());
 224                 }
 225         }
 226
 227         _analysis.set_samples_per_point (_samples_per_point);
 228         _analysis.set_sample_rate (_film->audio_frame_rate ());
 229         _analysis.set_leqm (_leqm->leq_m());
 230 }